Huge-Volume Low-Frequency Floating Car Data Map-Matching And Travel Time Estimation

The accurate and timely acquisition of traffic information is vital to urban transport management and planning. The collected traffic information can not only mitigate traffic congestions in the road network but also provide real-time route guidance services to public users. In recent years, the floating car system becomes increasing popular for collecting traffic information. This floating car system utilizes a large fleet of vehicles equipped with global positioning systems (GPS) and wireless communication devices. The collected trajectories of these floating cars can be a very useful data source for generating traffic information due to its low cost and large spatial coverage. This study focuses on low-frequent floating car data (FCD) generated by ten thousands of taxis in Wuhan city. In this study, an efficient and accurate algorithm is developed for matching huge-volume low-frequent FCD onto the road network. Then, a method is proposed to estimate accurately link travel time information based on low-frequent FCD. A real world case study is carried out to demonstrate the applicability of the proposed map matching algorithm and travel time estimation method. This study contributes to the literature in following four aspects:1. The pre-process technique of huge-volume low-frequent floating car data. Using FCD generated by12,000taxis in Wuhan, the characteristics of FCD are analyzed, including data format, instantaneous travel speed and direction, data volume collected in different time periods, data sampling frequency, and passenger loading status. For solving the network deviation error due to the encryption of road maps, a method is proposed to calibrate rasterized road network coordinate system and floating car coordinate system. The proposed method improves the calibration accuracy between these two coordinate systems, and thus facilitates the effective processing of huge-volume FCD.2. A primary map matching algorithm for huge-volume low-frequent FCD. Computational performance of matching huge-volume FCD onto the road network is one of critical factors for the FCD applications. The primary map matching is to match GPS points to the network links with a matching score less than a given threshold. Because FCD have a huge data volume (e.g. more than12,000taxis in Wuhan) and the road network contains ten thousands of links (e.g. more than 26,000links in Wuhan), computational efficiency of map matching algorithm can have a significant impact on the process of FCD. This study first analyzes the proposed map matching algorithm built on the rasterized road network, and discusses how to determine the optimal raster size of the road network. Then, a primary map matching algorithm based on rasterized road network is proposed in order to enhance the computational efficiency of matching huge-volume low frequent FCD.3. The study of low-frequent FCD trajectory recovery The effects of the number of GPS points in the trajectory recovery process are discussed. It is found that at least three GPS points are required for correctly recovering the trajectory of a floating car moving in the complex road network. The more number of GPS points we have, the more robust results will be obtained but the more computational resources are required. Then, the proposed map matching is optimized for scenarios when passenger loading status changes and the quality of FCD are unstable. The optimal strategy for selecting candidate nodes on the links within the route search area is developed to further enhance the computational efficiency of the proposed algorithm.4. Travel time estimation method using low-frequent FCD Travel time information is a key factor for evaluating the network performance. Based on the comprehensive review of existing methods, a new travel time estimation method is proposed by fully using the travel statuses of floating cars at downstream and upstream of road junctions. In the proposed method, turn delays at road junctions can be estimated accurately by using the vehicle location and instantaneous travel speed information respectively collected at downstream and upstream of road junctions.